Is there a better way to optimize IIR implementation in speech processing?

Hi, all
Nowadays I work on the optimization of speech processing,
but most of my previous work is about video codec, so
I have little experience about speech processing.

In speech processing, IIR filter is always employed. In IIR filter
data dependency makes optimization using SIMD difficult.
The following code is about IIR. I can't find a better way to
optimize it.

while( (Nx) > 0 )
{
Nx--;
z0 = (*x) - a1 * z1 - a2 * z2;
*(x++) = b0 * z0 + b1 * z1 + b2 * z2;
z2 = z1;
z1 = z0;
}
Can anyone give me some suggestions?

Thanks.

Best Regards
Jogging Start a New Thread